X-Band Outphasing Power Amplifier with InternalLoad Modulation Measurements

Michael Litchfield, Tibault Reveyrand and Zoya PopovicDepartment of Electrical, Computer, and Energy Engineering (ECEE)

University of Colorado at Boulder, CO 80309-0425, USA

Abstract—This paper presents a general measurement baseddesign method for outphasing amplifiers. Both isolated and non-isolated combiners are designed based on individual PA load-pull measurements, in order to load modulate two 0.15µmGaN MMIC power amplifiers at 10.1 GHz. Load modulationmeasurements are performed at ports internal to the PA, withthe inclusion of couplers in the combiner and a four-port LSNA,to gain insight into the dynamics of load modulation duringoutphasing operation. Both outphasing systems exhibit a peakPAE greater than 41.5 % at a peak output power greater than35.7 dBm.

Keywords—Power Amplifiers, Outphasing, Load Modulation,GaN, LSNA.

I. INTRODUCTION

Architectures using multiple power amplifiers, such asDoherty and outphasing, are commonly used to maintain highpower added efficiency (PAE) over a large output powerrange. An insightful analysis of internal PAs requires internalmeasurement with dedicated instrumentation. Conventionalmeasurement techniques focus on either externally measurableparameters of the PA system, or measurement of static subsys-tems. Such measurements fall short of predicting subsystemperformance in full system operation, especially under loadmodulation conditions. Multiple PA circuits can benefit frommeasurements at internal circuit nodes, such as the time-domain measurement of waveforms at the interstage of aPA with gate voltage shaping [1]. Individual performance ofthe main and peaking amplifiers in a Doherty PA were alsoreported using a non-contact near-field technique [2]. The inputwaveforms of a broadband Doherty-outphasing PA using a2-port LSNA are used to determine input phase, while theoutput is only measured with a power meter in [3]. The resultspresented here further the understanding of the outphasingPA through internal circuit measurements at both input andoutput of the internal PAs during full outphasing operation.In particular, v and i measurements in the output ports ofthe individual PAs give insight into the dynamic load-pullingmechanism of a non-isolated outphasing PA.

An outphasing amplifier [4] consists of two high-efficiencyPAs, each driven in parallel with constant input power, and anoutput power combiner that can be isolated or non-isolated.The differential phase of the input signal contains envelopeinformation that is reconstructed at the output. In non-isolatedoutphasing PAs, load modulation is significant and variesthe individual PA performances. Therefore, it is importantto understand and verify experimentally the effect of loadmodulation on the efficiency and power of each PA. Analysesleading to the design of non-isolated combiners in [5]–[7]

Fig. 1. The X-Band MMIC PA (a) used in this design is a single-stage PAusing a 10 × 100µm FET in TriQuint’s 0.15µm GaN process. The chip is3.8 mm × 2.3 mm. The MMIC is mounted in a 2.9 mm coaxial test-fixture(b), with internal measurements referenced to the RF pads of the MMIC. Thetest-fixture error terms are extracted from TRL calibration measurements.

reference the load modulation to the virtual drain of an idealvoltage source or ideal PA class, and focus on designing thecombining power factor. In a practical design, the PA and com-biner are often designed separately [8] and the load modulationshould be designed to optimize performance by referencing itto the PA output. This paper presents a measurement-basedmethod for designing X-Band outphasing power amplifiers.This method consists of two steps: first, the MMIC PAs aredesigned to maximize efficiency and characterized in largesignal for load modulation design, then hybrid combiners aredesigned to maximize system performance with the givenPAs. Both isolated and non-isolated combiners are designedbased on the large signal characterization of individual PAsat 10.1 GHz. The dynamic load modulation between the inter-nal PAs is measured to validate the design method. Finally,internal (individual PAs) and external system performancesare measured and presented for both combiner topologies. Tothe best of the authors’ knowledge, an outphasing PA has notbeen presented above 5 GHz [9] and internal load modulationmeasurements have not been studied to date.

II. MMIC PA DESIGN AND CHARACTERIZATION

This work utilizes two Class-B GaN MMIC PAs fabricatedin TriQuint’s 0.15µm process, as described in [10] and shownin Fig. 1. To inform the design of the hybrid combiner (loadmodulation), the MMIC PAs are first characterized with load-pull measurements.

The large-signal measurement setup, seen in Fig. 2, is basedon a large signal network analyzer (LSNA): the VTD SWAPX-402 [11], which can measure the RF voltage and currentat a defined reference plane after calibration. The system iscalibrated with an absolute coaxial SOLT method [12]. TheSWAP does not include any Harmonic Phase Reference (HPR).The absolute phase calibration is performed during the reverse”Thru” measurement by one of the LSNA receivers (seen

978-2-87487-035-4 © 2014 EuMA 6-9 Oct 2014, Rome, Italy

Proceedings of the 44th European Microwave Conference

1428

CouplerCoupler

Driverf0

SWAP

10M

Hz

Ref

. 1 2Passive Tuner

MMICPA

Fig. 2. The 2-port LSNA measurement setup for MMIC PA load-pullcharacterization. Measurement reference planes are de-embedded to the MMICPA RF pads within the test-fixture.

as a RF scope), which is connected directly to the RF pathas explained in [13]. After calibration, system measurementsare associated with an error-term matrix defined in Eqn.(1), establishing a correction to apply at the measured RFfrequency between the reference plane ( 1© and 2© in Fig. 2),measured quantities (RF voltages vi and current ii), and theraw data acquired at the image frequency on the LSNA’s ADCs(noted ri). v1i1v2

i2

=

α1 β1 0 0γ1 δ1 0 00 0 α2 β20 0 γ2 δ2

.r1r2r3r4

(1)

Each LSNA port is associated with a unique, 4-term errormatrix. Internal measurements are de-embedded to the RF padson the GaN MMIC PA, as seen in Fig. 1, using the test-fixtureerror terms extracted from a TRL calibration measurement.

Load-pull contours (PAE and Pout) are summarized inFig. 3, along with designed load modulation contours (ΓPA1

and ΓPA2) which are discussed in the next section. At a drainvoltage of 20 V, the measured peak power is 35.5 dBm andthe measured peak PAE is 67.1 %. A critical part of PA char-acterization for outphasing amplifier design is identifying theaxis of impedances over which the power load-pull contoursare symmetric. The load modulation (ΓPA1 and ΓPA2) of thecombiner is designed to be symmetric over this axis to ensurethat each PA produces equal output power for stability andbalance.

III. OUTPHASING COMBINER DESIGNS

In order to measure internally the individual PA perfor-mances and dynamic load modulation, it is necessary to in-clude low-loss ( 0.2 dB) bi-directional couplers in the combinerdesign, as seen in Fig. 4(top). Both isolated and non-isolatedcombiners are designed, fabricated, and tested as follows.

A. Non-Isolated Combiner

The non-isolated combiner is shown in Fig. 4a and fabri-cated on 30 mil Rogers 4350B substrate. The design is basedon a tee-junction topology and optimized by adding conjugatesusceptances [4] and varying the lengths of the tee-junctionlines around 90 to adjust the load modulation [14]. An S-parameter analysis, at the fundamental frequency, on the non-isolated combiner yields equations useful for designing the

0 1.0

1.0

10.0

10.0

5.0

5.0

2.0

2.0

3.0

3.0

4.0

4.0

0.2

0.2

0.4

0.4

0.6

0.6

0.8

0.8

Fig. 3. Measured output power and PAE load-pull contours at a constant inputpower (Pin = 26 dBm) with axis of power symmetry shown for design, andload modulation predicted by Eqns. (3), (4) for non-isolated combiner. CWload-pull measurements have been performed with a LSNA (VTD-SWAP) anda passive tuner at 10.1 GHz. PAE contours are shown from 30 % to 60 % (witha 10 % step) and Pout contours are traced from 28 to 35 dBm (with 1 dB step).

a1

b1

a3 b3a2

b2(b)(a)

4PA

PA

3

21 OutputCombiner (a) or (b)

ΓPA1

ΓPA2

θ

θ

Fig. 4. Principle of outphasing PA illustrated on the top. The power combinercan be non-isolated (a) or isolated (b). The fabricated combiners are operatingat 10.1 GHz, on 30 mil Ro4350B substrate.

load modulation contours. Defining the input power wavesshown in Fig. 4a as:

a1 = x · a2ejθ, a2 =1

x· a1e−jθ, a3 = 0 (2)

where the port 3 is assumed to be matched. x predicts the effectof PA output power imbalance, and θ is the swept differentialphase between the branches. Equations for b1 and b2 can besolved for the reflection coefficient at each input port of thecombiner, corresponding to the load modulation at the outputof each individual PA, as:

ΓPA1 = S11 +1

x· S12e

−jθ (3)

ΓPA2 = S22 + x · S21ejθ (4)

The predicted load modulation is shown in Fig. 3. Becausethe load-pull characterization is performed at the MMIC ref-erence plane, the design of the load modulation should occurat this plane as well. All transitions (including test-fixture)

1429

Low-Loss Coupler

Low-Loss CouplerMMIC

RF Switches

Com

bine

r

Pow

erM

eter

Coupler

Driverf0

Driver

f0 φ Coupler

SWAP

10M

Hz

Ref

.

PA

MMICPA

1 2

3 4

Fig. 5. The measurement setup is based on a 4-port LSNA. The two outputcouplers are included in the combiner, enabling internal load modulationmeasurements. A phase shifter (φ) sweeps the differential phase at an inputpower level which saturates both PAs.

and couplers on the output of the MMIC are measured andincluded in the combiner design. The design goal is to intersectthe load modulation contours at the measured peak efficiencyand remain in the highest efficiency impedance region possible,while maintaining symmetry over the axis of power symmetry.This design reduces the output power in order to ensure thePAs are loaded for highest possible efficiency near peak outputpower.

B. Isolated Combiner

The isolated combiner, a standard rat-race circuit depictedin Fig. 4b, provides a 180 phase shift in order to sum anddifference the signals from the parallel PAs. The loss betweeneither PA input and either output is between 1.0 and 1.4 dB,mostly due to the losses in the test-fixture and couplers. Thereturn loss on both input ports is better than 19.5 dB, andthe isolation between the PA inputs is 22.5 dB. The measuredphase between paths to the summing output (∠S31−∠S32) is4.5, while the phase between paths to the difference output(∠S41 − ∠S42) is 173.

IV. INTERNAL LOAD MODULATION MEASUREMENTS

The large-signal measurement setup presented in section IIhas been enhanced to 4-port capability with a switch-matrixconsisting of four RF SPDTs as shown in Fig. 5. One ofthe four ports (noted 1©) is directly connected to two LSNAreceiver channels, providing a phase reference for the threeother ports, which are measured sequentially and are timedomain-aligned to the first port.

The calibration of the system consists of performing threestandard LSNA calibrations, one for each switch matrix po-sition, in order to extract an error-term calibration matrix forevery port as follows:(

viii

)=

[αi βiγi δi

].

(r3r4

)(5)

with i = 2, 3 and 4 for ports 2, 3, and 4 respectively. Each portis calibrated with the usual Short, Open, and Load standards,

Min power (θ = 180)

Peak Power (θ = 6.5)

ΓPA1

ΓPA2

(a)

ΓPA1

ΓPA2

Peak PAEPeak Pout

(θ = 0)

(b)

Fig. 6. Measurements of load modulation seen by internal PAs with non-isolated (a) and isolated (b) combiners. The load impedance seen by eachPA is varied with the outphasing angle which changes the differential phasedriving the parallel PAs.

before being connected with the Thru standard to the port 1phase reference.

After calibration, the system measures RF voltage andcurrent waveforms at the input and output of the two individualMMIC PAs. The complete outphasing PA system performanceis measured simultaneously with a power meter at the outputof the combiner as in [3]. During measurement, RF voltagesand currents at ports 2, 3 and 4 are acquired sequentially witha common port 1 measurement. A time domain alignment isperformed by setting the phase of v1 to 0 at the fundamentalfrequency. Taking the input voltage as a phase reference is notan issue in a outphasing setup because the input power levelremains constant and large.

A phase shifter (φ) sweeps the phase difference between thetwo branches. The second branch source amplitude is adjustedto maintain amplitude balance at the two inputs. Internal PAand load modulation measurements are de-embedded to theMMIC PA RF pads, shown in Fig. 1.

A. Load Modulation Measurement Results

For each phase point θ = ∠Vin 2(f0)/Vin 1(f0), the loadimpedance seen by each power amplifier was measured bythe SWAP. This provides a measurement of the designed loadmodulation contours.

Fig. 6a shows the measured load seen by each PA as thephase between the two input voltages is swept from −180

to 180. This measurement agrees with the designed loadmodulation shown in Fig. 3. The peak output power occursnear the crossing of the two load contours at an outphasingangle of 6.5, due to their combined proximity to the peakpower impedance. The peak PAE is measured at θ = 10.3.The minimum measured output power is 3.6 dBm, reached asthe two load contours approach the edge of the Smith Chart.

Load modulation should not occur for the ideal isolatedoutphasing combiner with infinite isolation and 50 Ω inputimpedances at the driven ports. Fig. 6b shows the measuredload modulation for the designed rat-race isolated combiner.The small load modulation circles are expected, because theinput impedances at the driven ports are not exactly 50 Ω at theMMIC reference plane and the isolation is finite. Furthermore,one load modulation circle is larger than the other because ofa small internal PA output power imbalance.

1430

180 120 60 0 60 120 180- - -

706050403020100

706050403020100

(b)

(a)

Fig. 7. Measured non-isolated (a) and isolated (b) outphasing systemoutput powers and efficiencies and internal PA efficiencies. The non-isolatedcombiner shows a maximum PAE of 41.5 % at θ = 10.3 and a maximumPout of 35.7 dBm at θ = 6.5. The isolated combiner shows a peak PAE of41.6 % at θ = 8.2 and a maximum Pout of 35.8 dBm at θ = 3.7.

V. OUTPHASING SYSTEM PERFORMANCE

The system performance of the non-isolated outphasing PAis shown in Fig. 7a. The design achieves a peak output powerof 35.7 dBm with a peak PAE (PAEtotal) of 41.5 %. Thisincludes the loss in the couplers and transitions on the output.In an integrated design without couplers, the loss would reduceby 1.3 dB, increasing the peak output power to 37 dBm and thepeak PAE to 60 %. The PAE of the internal PAs is shown inFig. 7a: PAEPA1 and PAEPA2. Because the PAs are notideal or balanced, maximum internal PAEs do not occur at thesame outphasing angle as maximum system efficiency.

Fig. 7b shows the system performance of the isolatedoutphasing amplifier. As expected, the system efficiency isalmost exactly symmetric about the peak power which occursvery close to zero outphasing angle. The measured peak poweris 35.8 dBm and the peak PAE is 41.6 %. Again, losses incouplers and transitions are included, so integration woulddecrease the loss by 1 dB, increasing the peak power to 36.8 dBand peak PAE to 59.1 %. Because the PAs only experience asmall range of loading due to the combiner isolation, theirindividual PAEs are flat across outphasing angle. Thus thesystem PAE is most influenced by the system output powercharacteristic, which causes the PAE roll-off with power back-off level to be steep.

Finally, Fig. 8 compares the system efficiencies for bothmeasured systems above. The PAE is plotted against the powerback-off level, relevant to high PAR signals. Note that systemdrain efficiencies are 10 % greater than PAEs, due to the use ofcompressed, single-stage individual PAs as well as the constantinput power in the outphasing system.

VI. CONCLUSION

A practical design methodology that treats both the PA andcombiner is demonstrated through implementation at 10.1 GHzwith GaN MMIC PAs and hybrid combiners. The internal

0

10

20

30

40

50

08- 6- 4- 2-

Fig. 8. Comparison of measured system PAE’s for isolated and non-isolatedoutphasing.

load modulation is measured to confirm the design and giveinsight into internal behavior during dynamic operation. Dueto care taken to balance the internal PA output powers, themeasured load modulation is well-behaved and close to theexpected design. Both isolated and non-isolated outphasingPAs are demonstrated with peak PAEs higher than 41 % andpeak output powers above 35.7 dBm (PAEs higher than 59 %and peak output powers above 5 W for integrated designs).

REFERENCES

[1] A. Ramadan et al., “Two-stage gan hemt amplifier with gate-sourcevoltage shaping for efficiency versus bandwidth enhancements,” IEEETransactions on MTT, vol. 59, no. 3, pp. 699–706, 2011.

[2] R. Hou et al., “Non-intrusive characterization of active device interac-tions in high-efficiency power amplifiers,” in IEEE MTT-S IMS Digest.IEEE, 2013, pp. 1–3.

[3] C. Andersson et al., “A 1-3-ghz digitally controlled dual-rf input power-amplifier design based on a doherty-outphasing continuum analysis,”IEEE Transactions on MTT, vol. 61, no. 10, pp. 3743–3752, 2013.

[4] H. Chireix, “High power outphasing modulation,” Radio Engineers,Proceedings of the Institute of, vol. 23, no. 11, pp. 1370–1392, 1935.

[5] F. Raab, “Efficiency of outphasing rf power-amplifier systems,” IEEETransactions on Communications, vol. 33, no. 10, pp. 1094–1099, 1985.

[6] M. El-Asmar et al., “Analytical design methodology of outphasingamplification systems using a new simplified chireix combiner model,”IEEE Transactions on MTT, vol. 60, no. 6, pp. 1886–1895, 2012.

[7] B. Stengel and W. Eisenstadt, “Linc power amplifier combiner methodefficiency optimization,” Vehicular Technology, IEEE Transactions on,vol. 49, no. 1, pp. 229–234, 2000.

[8] D. Calvillo-Cortes et al., “A package-integrated chireix outphasingrf switch-mode high-power amplifier,” IEEE Transactions on MTT,vol. 10, no. 61, pp. 3721 – 3732, 2013.

[9] J. Grundlingh, K. Parker, and G. Rabjohn, “A high efficiency Chireixout-phasing power amplifier for 5 GHz WLAN applications,” in IEEEMTT-S IMS Digest, vol. 3. IEEE, 2004, pp. 1535–1538.

[10] S. Schafer et al., “X-Band MMIC GaN Power Amplifiers Designed forHigh-Efficiency Supply-Modulated Transmitters,” in IEEE MTT-S IMSDigest, Seattle, WA, 2013, pp. 1–4.

[11] P. Roblin, Nonlinear RF circuits and nonlinear vector network analyz-ers. Cambridge University Press, 2011.

[12] J. Verspecht, “Calibration of a measurement system for high frequencynonlinear devices,” Brussels, Belgium Vrije Univeriteit Brussels, Ph. D.Thesis, 1995.

[13] K. Andersson and C. Fager, “Oscilloscope based two-port measurementsystem using error-corrected modulated signals,” in Integrated Nonlin-ear Microwave and Millimetre-Wave Circuits (INMMIC). IEEE, 2012,pp. 1–3.

[14] W. Gerhard and R. Knoechel, “Improved design of outphasing poweramplifier combiners,” in German Microwave Conference, 2009, pp. 1–4.

1431